Fundamentals of Experiment Design - Basic Principles in Experimental Design

Basic Principles in Experimental Design

A good experimental design should be effective, manageable, efficient, and capable of being monitored, controlled, and evaluated. Effectiveness relates to the ability to achieve the planned or outlined objectives, targets, and usefulness. Manageability relates to various limitations or constraints in conducting experiments or in analyzing data. Efficiency means rationalization in the use of resources, funds, and time to obtain information from the experiment.

Experimental design deals with techniques for overcoming and controlling variability or variables that disturb the actual influence of the treatment or factor under study, referred to as Environmental Design.

There are two types of sources of variability in experimental design:

• Main factors are the factors that will be studied and deliberately given.
• External factors: factors whose influence is expected to be as small as possible. These factors consist of:
  • Factors that can be identified and their influence estimated before the experiment. For instance, in the case of wanting to know the difference between two corn varieties, if the two varieties give different results, the difference in these results, in addition to being caused by the difference in varieties, may also be caused by differences in soil fertility. To overcome this, grouping is usually performed, so that the variability among groups can be measured and removed from the experimental error.
  • Factors that can be identified but whose influence cannot be anticipated. For example, in the case above, if the field has a gradual fertility direction from left to right so that the yield will decrease from left to right, if variety A is always planted to the right of variety B, then in this case variety B will be advantaged because relatively it is located on a more fertile land than variety A. So, in this case, the appearance of the yield of varieties A and B will be biased and favor B, and if we want to compare varieties A and B, the difference that occurs is not solely caused by the difference in varieties but also caused by differences in soil fertility. To overcome this, randomization is performed.
  • Factors that cannot be identified. To overcome this, replication is performed.

To minimize experimental error to improve the accuracy of the experiment, it is necessary to have replication, randomization, and local environment control, which are the basic principles in experimental design. Orthogonality, confounding, and efficiency are additional principles. Other principles include orthogonality, confounding, and efficiency.

Local Environment Control

Local environment control means controlling environmental conditions that have the potential to affect the response to treatment. This can be done by:

• Experimental design. This is usually done by grouping the experimental units and each group contains all treatments so that the variability within the group is minimized and the variability between groups is maximized.
  

  Figure 1.1 Example of grouping experimental plots

• Use of covariates. This is done if there is variability among experimental units. For example, if you want to know the difference in the effect of certain types of feed on the weight gain of chickens. Before being fed, the chickens already have different weights, so a covariate, the weight of the chicken before feeding, is used. Analysis using this variable in statistics is known as analysis of covariance.
• Increasing the size of the experimental unit. The smaller the experimental error (), the greater the information obtained (or ). In other words, the larger the size of the experimental unit (n), the smaller the experimental error and the greater the information.

Randomization

Randomization is carried out by giving an equal chance to each experimental unit to be subjected to treatment. Randomization is also used to eliminate bias. For example, in an experiment with two corn varieties, the field has a fertility direction gradually from left to right so that the yield will decrease from left to right. To avoid this bias, the varieties are placed randomly on the experimental plots.

Figure 1.2 Example of systematic / non-random placement of plots

The non-random placement of the plots does not provide a valid estimate of the experimental error and will give biased results. In the example above, the field has a fertility direction gradually from left to right so that the yield will decrease from left to right. If variety A is always planted to the left of variety B, then in this case variety A will be advantaged because relatively treatment A is located on more fertile land than variety B. So in this case, the appearance of the yield of varieties A and B will be biased and favor A, and if we want to compare varieties A and B, the difference that occurs is not solely caused by the difference in varieties but also caused by differences in soil fertility. To avoid this, the plots must be placed in such a way that no variety is advantaged or disadvantaged. This can be done by placing the varieties randomly on the experimental plots.

Repetition

Repetition involves applying the same treatment to the experimental unit more than once. The function of repetition includes:

• Estimation of error. If an experiment does not contain repetitions, the experimental error cannot be estimated. We cannot explain precisely whether the differences that arise are due to differences among treatments or differences among experimental units.
• Increasing experimental precision. The use of less precise techniques or the use of less homogeneous experimental units can be overcome by increasing the number of repetitions. With an increase in repetitions, the population mean estimate will be more precise.
• Expanding the scope of conclusions. This is done through the selection of more varied experimental units, for example, repetitions done at different times.
• Controlling error variance. By making groups as repetitions, the experimental units within the group have minimum diversity, and the experimental units between groups have maximum diversity, making efforts to see treatment differences within the group more precise. This way, error variance can be controlled.

In determining the number of repetitions, several things that need to be considered include the variability of tools, materials, media, and the experimental environment, as well as the cost and labor available.

Each principle in experimental design plays an important role in ensuring the validity and reliability of the experiment's results. With these principles, we can ensure that our experimental results are accurate and trustworthy, and minimize the influence of factors we do not want.